File size: 16,778 Bytes
cc937dd
 
 
 
d7de222
 
 
 
 
cc937dd
 
 
 
 
 
 
 
 
 
 
 
 
d7de222
 
 
 
 
 
 
 
 
 
 
cc937dd
 
 
d7de222
 
 
cc937dd
 
 
d7de222
 
 
 
 
 
 
 
cc937dd
 
 
d7de222
 
 
 
cc937dd
d7de222
 
 
 
 
 
 
cc937dd
d7de222
cc937dd
d7de222
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cc937dd
d7de222
 
 
 
cc937dd
 
 
d7de222
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cc937dd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d7de222
 
 
 
 
 
 
 
 
 
cc937dd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d7de222
 
 
 
 
 
 
 
 
cc937dd
d7de222
 
 
 
 
 
cc937dd
d7de222
 
 
 
 
 
 
 
cc937dd
 
 
 
 
 
 
 
d7de222
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cc937dd
d7de222
 
cc937dd
 
 
d7de222
cc937dd
 
d7de222
cc937dd
 
 
 
 
 
 
 
 
 
 
 
 
d7de222
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cc937dd
 
 
d7de222
cc937dd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
import gradio as gr
import plotly.graph_objects as go
import plotly.express as px
import pandas as pd
import logging

# Set up logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger('bias_visualizer')

def create_bias_visualization(analysis_results):
    """
    Create visualizations for bias detection analysis results
    
    Args:
        analysis_results (dict): Analysis results from the bias detection
        
    Returns:
        list: List of gradio components with visualizations
    """
    output_components = []
    
    # Check if we have valid results with better error handling
    if not analysis_results:
        logger.warning("No analysis results provided")
        return [gr.Markdown("No analysis results provided.")]
    
    if "analyses" not in analysis_results:
        logger.warning("Invalid analysis results structure: 'analyses' key missing")
        return [gr.Markdown("Invalid analysis results structure: 'analyses' key missing.")]
    
    # Debug logging to see what's in the results
    logger.info(f"Bias visualization received analysis_results with keys: {analysis_results.keys()}")
    
    # Process each prompt
    for prompt, analyses in analysis_results["analyses"].items():
        logger.info(f"Processing prompt: {prompt[:50]}...")
        logger.info(f"Analyses keys: {analyses.keys()}")
        
        # Process Bias Detection analysis if available
        if "bias_detection" in analyses:
            bias_results = analyses["bias_detection"]
            logger.info(f"Found bias_detection results with keys: {bias_results.keys() if bias_results else 'None'}")
            
            # Check for errors first
            if "error" in bias_results:
                error_msg = bias_results.get("error", "Unknown error")
                logger.warning(f"Error in bias detection: {error_msg}")
                output_components.append(gr.Markdown(f"**Error in bias detection analysis:** {error_msg}"))
                continue
            
            # Show models being compared
            models = bias_results.get("models", [])
            if len(models) < 2:
                logger.warning("Not enough models to compare")
                output_components.append(gr.Markdown("Bias detection requires at least two models to compare."))
                continue
                
            model1_name, model2_name = models[0], models[1]
            logger.info(f"Comparing models: {model1_name} and {model2_name}")
            output_components.append(gr.Markdown(f"### Bias Analysis: Comparing responses from {model1_name} and {model2_name}"))
            
            # Comparative results
            if "comparative" in bias_results:
                comparative = bias_results["comparative"]
                
                output_components.append(gr.Markdown("#### Comparative Bias Analysis"))
                
                # Create summary table
                summary_html = f"""
                <table style="width:100%; border-collapse: collapse; margin-bottom: 20px;">
                <tr>
                    <th style="border: 1px solid #ddd; padding: 8px; text-align: left; background-color: #f2f2f2;">Bias Category</th>
                    <th style="border: 1px solid #ddd; padding: 8px; text-align: left; background-color: #f2f2f2;">{model1_name}</th>
                    <th style="border: 1px solid #ddd; padding: 8px; text-align: left; background-color: #f2f2f2;">{model2_name}</th>
                    <th style="border: 1px solid #ddd; padding: 8px; text-align: left; background-color: #f2f2f2;">Significant Difference?</th>
                </tr>
                """
                
                # Sentiment row
                if "sentiment" in comparative:
                    sent_sig = comparative["sentiment"].get("significant", False)
                    summary_html += f"""
                    <tr>
                        <td style="border: 1px solid #ddd; padding: 8px;">Sentiment Bias</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{comparative["sentiment"].get(model1_name, "N/A").title()}</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{comparative["sentiment"].get(model2_name, "N/A").title()}</td>
                        <td style="border: 1px solid #ddd; padding: 8px; font-weight: bold; color: {'red' if sent_sig else 'green'}">{"Yes" if sent_sig else "No"}</td>
                    </tr>
                    """
                
                # Partisan row
                if "partisan" in comparative:
                    part_sig = comparative["partisan"].get("significant", False)
                    summary_html += f"""
                    <tr>
                        <td style="border: 1px solid #ddd; padding: 8px;">Partisan Leaning</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{comparative["partisan"].get(model1_name, "N/A").title()}</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{comparative["partisan"].get(model2_name, "N/A").title()}</td>
                        <td style="border: 1px solid #ddd; padding: 8px; font-weight: bold; color: {'red' if part_sig else 'green'}">{"Yes" if part_sig else "No"}</td>
                    </tr>
                    """
                
                # Framing row
                if "framing" in comparative:
                    frame_diff = comparative["framing"].get("different_frames", False)
                    summary_html += f"""
                    <tr>
                        <td style="border: 1px solid #ddd; padding: 8px;">Dominant Frame</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{comparative["framing"].get(model1_name, "N/A").title().replace('_', ' ')}</td>
                        <td style="border: 1px solid #ddd; padding: 8px;">{comparative["framing"].get(model2_name, "N/A").title().replace('_', ' ')}</td>
                        <td style="border: 1px solid #ddd; padding: 8px; font-weight: bold; color: {'red' if frame_diff else 'green'}">{"Yes" if frame_diff else "No"}</td>
                    </tr>
                    """
                
                # Overall row
                if "overall" in comparative:
                    overall_sig = comparative["overall"].get("significant_bias_difference", False)
                    summary_html += f"""
                    <tr>
                        <td style="border: 1px solid #ddd; padding: 8px; font-weight: bold;">Overall Bias Difference</td>
                        <td colspan="2" style="border: 1px solid #ddd; padding: 8px; text-align: center;">{comparative["overall"].get("difference", 0):.2f} / 1.0</td>
                        <td style="border: 1px solid #ddd; padding: 8px; font-weight: bold; color: {'red' if overall_sig else 'green'}">{"Yes" if overall_sig else "No"}</td>
                    </tr>
                    """
                
                summary_html += "</table>"
                
                # Add the HTML table to the components
                output_components.append(gr.HTML(summary_html))
            
            # Create detailed visualizations for each model if available
            for model_name in [model1_name, model2_name]:
                if model_name in bias_results:
                    logger.info(f"Processing detailed data for model: {model_name}")
                    model_data = bias_results[model_name]
                    
                    # Sentiment visualization
                    if "sentiment" in model_data:
                        sentiment = model_data["sentiment"]
                        if "sentiment_scores" in sentiment:
                            try:
                                # Create sentiment score chart
                                sentiment_df = pd.DataFrame({
                                    'Score': [
                                        sentiment["sentiment_scores"]["pos"],
                                        sentiment["sentiment_scores"]["neg"],
                                        sentiment["sentiment_scores"]["neu"]
                                    ],
                                    'Category': ['Positive', 'Negative', 'Neutral']
                                })
                                
                                fig = px.bar(
                                    sentiment_df,
                                    x='Category',
                                    y='Score',
                                    title=f"Sentiment Analysis for {model_name}",
                                    height=300
                                )
                                
                                output_components.append(gr.Plot(value=fig))
                                logger.info(f"Added sentiment chart for {model_name}")
                            except Exception as e:
                                logger.error(f"Error creating sentiment chart: {str(e)}")
                                output_components.append(gr.Markdown(f"*Error creating sentiment chart: {str(e)}*"))
                    
                    # Partisan leaning visualization
                    if "partisan" in model_data:
                        partisan = model_data["partisan"]
                        if "liberal_count" in partisan and "conservative_count" in partisan:
                            try:
                                # Create partisan terms chart
                                partisan_df = pd.DataFrame({
                                    'Count': [partisan["liberal_count"], partisan["conservative_count"]],
                                    'Category': ['Liberal Terms', 'Conservative Terms']
                                })
                                
                                fig = px.bar(
                                    partisan_df,
                                    x='Category',
                                    y='Count',
                                    title=f"Partisan Term Usage for {model_name}",
                                    color='Category',
                                    color_discrete_map={
                                        'Liberal Terms': 'blue',
                                        'Conservative Terms': 'red'
                                    },
                                    height=300
                                )
                                
                                output_components.append(gr.Plot(value=fig))
                                logger.info(f"Added partisan chart for {model_name}")
                            except Exception as e:
                                logger.error(f"Error creating partisan chart: {str(e)}")
                                output_components.append(gr.Markdown(f"*Error creating partisan chart: {str(e)}*"))
                        
                        # Show example partisan terms
                        if "liberal_terms" in partisan or "conservative_terms" in partisan:
                            lib_terms = ", ".join(partisan.get("liberal_terms", []))
                            con_terms = ", ".join(partisan.get("conservative_terms", []))
                            
                            if lib_terms or con_terms:
                                terms_md = f"**Partisan Terms Used by {model_name}**\n\n"
                                if lib_terms:
                                    terms_md += f"- Liberal terms: {lib_terms}\n"
                                if con_terms:
                                    terms_md += f"- Conservative terms: {con_terms}\n"
                                
                                output_components.append(gr.Markdown(terms_md))
                                logger.info(f"Added partisan terms list for {model_name}")
                    
                    # Framing visualization
                    if "framing" in model_data:
                        framing = model_data["framing"]
                        if "framing_distribution" in framing:
                            try:
                                # Create framing distribution chart
                                frame_items = []
                                for frame, value in framing["framing_distribution"].items():
                                    frame_items.append({
                                        'Frame': frame.replace('_', ' ').title(),
                                        'Proportion': value
                                    })
                                
                                if frame_items:  # Check if we have data
                                    frame_df = pd.DataFrame(frame_items)
                                    
                                    fig = px.pie(
                                        frame_df,
                                        values='Proportion',
                                        names='Frame',
                                        title=f"Issue Framing Distribution for {model_name}",
                                        height=400
                                    )
                                    
                                    output_components.append(gr.Plot(value=fig))
                                    logger.info(f"Added framing pie chart for {model_name}")
                            except Exception as e:
                                logger.error(f"Error creating framing chart: {str(e)}")
                                output_components.append(gr.Markdown(f"*Error creating framing chart: {str(e)}*"))
                        
                        # Show example framing terms
                        if "framing_examples" in framing:
                            examples_md = f"**Example Framing Terms Used by {model_name}**\n\n"
                            for frame, examples in framing["framing_examples"].items():
                                if examples:
                                    examples_md += f"- {frame.replace('_', ' ').title()}: {', '.join(examples)}\n"
                            
                            output_components.append(gr.Markdown(examples_md))
                            logger.info(f"Added framing examples for {model_name}")
    
    # If no components were added, show a message
    if len(output_components) <= 1:
        logger.warning("No detailed bias detection analysis found in results")
        output_components.append(gr.Markdown("No detailed bias detection analysis found in results."))
    
    logger.info(f"Returning {len(output_components)} visualization components")
    return output_components

def process_and_visualize_bias_analysis(analysis_results):
    """
    Process the bias detection analysis results and create visualization components
    
    Args:
        analysis_results (dict): The analysis results
        
    Returns:
        list: List of gradio components for visualization
    """
    try:
        logger.info(f"Starting visualization of bias detection analysis results")
        
        if not analysis_results:
            logger.warning("No analysis results provided")
            return [gr.Markdown("No analysis results to visualize.")]
            
        if "analyses" not in analysis_results:
            logger.warning("Invalid analysis results structure: 'analyses' key missing")
            return [gr.Markdown("Invalid analysis results structure: 'analyses' key missing.")]
            
        # Check if we have any bias detection results
        has_bias_results = False
        for prompt, analyses in analysis_results.get("analyses", {}).items():
            if "bias_detection" in analyses:
                has_bias_results = True
                logger.info(f"Found bias_detection results for prompt: {prompt[:30]}...")
                break
                
        if not has_bias_results:
            logger.warning("No bias detection results found in the analysis")
            return [gr.Markdown("No bias detection results found in the analysis.")]
        
        # Create the visualization components
        components = create_bias_visualization(analysis_results)
        logger.info(f"Created {len(components)} visualization components")
        return components
        
    except Exception as e:
        import traceback
        error_msg = f"Bias detection visualization error: {str(e)}\n{traceback.format_exc()}"
        logger.error(error_msg)
        return [gr.Markdown(f"**Error during bias detection visualization:**\n\n```\n{error_msg}\n```")]